Skip to content

ikod/cachetools

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Build Status codecov.io Dub downloads

cachetools

This package contains some cache implementations (for example LRU cache) and underlying data structures.

Why you may want to use it? Because it is fast, @safe. It is also @nogc and nothrow (inherited from your key/value types).

Limitations:

  • Cache implementations are not inherited from inerface or base class. This is because inheritance and attribute inference don't work together.

2Q cache

2Q cache is variant of multi-level LRU cache. Original paper http://www.vldb.org/conf/1994/P439.PDF It is adaptive, scan-resistant and can give more hits than plain LRU.

This cache consists from three parts (In, Out and Main) where 'In' receive all new elements, 'Out' receives all overflows from 'In', and 'Main' is LRU cache which hold all long-lived data.)

LRU cache

LRU cache keep limited number of items in memory. When adding new item to already full cache we have to evict some items. Eviction candidates are selected first from expired items (using per-cache configurable TTL) or from oldest accessed items.

Code examples

    auto cache = new CacheLRU!(int, string); // can be Cache2Q!(int, string)
    cache.size = 2048;      // keep 2048 elements in cache
    cache.ttl = 60.seconds; // set 60 seconds TTL for items in cache
    
    cache.put(1, "one");
    auto v = cache.get(1);
    assert(v == "one");  // 1 is in cache
    v = cache.get(2);
    assert(v.isNull);    // no such item in cache

Default values for TTL is 0 which means - no TTL. Default value for size is 1024;

Class instance as key

To use class as key with this code, you have to define toHash and opEquals(important: opEquals to the class instance not Object) as safe or trusted (optionally as nogc if you need it):

    import cachetools.hash: hash_function;
    class C
    {
        int s;
        this(int v)
        {
            s = v;
        }
        override hash_t toHash() const
        {
            return hash_function(s);
        }
        bool opEquals(const C other) pure const @safe
        {
            return s == other.s;
        }
    }
    CacheLRU!(immutable C, string) cache = new CacheLRU!(immutable C, string);
    immutable C s1 = new immutable C(1);
    cache.put(s1, "one");
    auto s11 = cache.get(s1);
    assert(s11 == "one");

Cache events

Sometimes you have to know if items are purged from cache or modified. You can configure cache to report such events. Important warning - if you enable cache events and do not check it after cache operations, then list of stored events will grow without bounds. Code sample:

    auto lru = new CacheLRU!(int, string);
    lru.enableCacheEvents();
    lru.put(1, "one");
    lru.put(1, "next one");
    assert(lru.get(1) == "next one");
    auto events = lru.cacheEvents();
    writeln(events);

output:

[CacheEvent!(int, string)(Updated, 1, "one")]

Each CacheEvent have key and val members and name of the event(Removed, Expired, Updated, Evicted). Events generated in case:

  • Event.Removed - when item removed from the cache via remove(key) or clear().
  • Event.Expired - when item has TTL and get(key) for this item called.
  • Event.Updated - when you call put(key, value) and key already presented in cache.
  • Event.Evicted - when cache have some limits (like total number of items in cache or total size) and this limit reached, so we have to purge something from cache.

Hash Table

Some parts of this package are based on internal hash table which can be used independently. It is open-addressing hash table with keys and values stored inline in the buckets array to avoid unnecessary allocations and better use of CPU cache for small key/value types.

Hash Table supports immutable keys and values. Due to language limitations you can't use structs with immutable/const members.

All hash table code is @safe and require from user supplied functions such as toHash or opEquals also be safe (or trusted).

It is also @nogc if toHash and opEquals are @nogc. opIndex is not @nogc as it can throw exception.

Several code samples:

import cachetools.containers.hashmap;

string[] words = ["hello", "my", "friend", "hello"];

void main()
{
    HashMap!(string, int) counter;

    build0(counter); // build table (verbose variant)
    report(counter);

    counter.clear(); // clear table

    build1(counter); // build table (less verbose variant)
    report(counter);
}

/// verbose variant
void build0(ref HashMap!(string, int) counter) @safe @nogc
{
    foreach(word; words)
    {
        auto w = word in counter;
        if ( w !is null )
        {
            (*w)++; // update
        }
        else
        {
            counter[word] = 1; // create
        }
    }
}
/// short variant
void build1(ref HashMap!(string, int) counter) @safe @nogc
{
    foreach(word; words)
    {
        auto w = word in counter;
        counter.getOrAdd(word, 0)++;
    }
}

void report(ref HashMap!(string, int) hashmap) @safe
{
    import std.stdio;
    writefln("keys: %s", hashmap.byKey);
    writefln("values: %s", hashmap.byValue);
    writefln("pairs: %s", hashmap.byPair);
    writeln("---");
}

Output:

keys: ["hello", "friend", "my"]
values: [2, 1, 1]
pairs: [Tuple!(string, "key", int, "value")("hello", 2), Tuple!(string, "key", int, "value")("friend", 1), Tuple!(string, "key", int, "value")("my", 1)]
---
keys: ["hello", "friend", "my"]
values: [2, 1, 1]
pairs: [Tuple!(string, "key", int, "value")("hello", 2), Tuple!(string, "key", int, "value")("friend", 1), Tuple!(string, "key", int, "value")("my", 1)]
---